A two-phase method based on virtual test samples and face recognition experiments

Abstract

In this paper, virtual test samples are exploited to perform face recognition. There are some basic ideas of the method. First, new test samples would be generated by using the symmetry of the original test samples. Second, samples used for training are exploited to represent original test samples and virtual test samples respectively. Deviation between these two kinds of samples (test/training) mentioned above is designed to find the M-nearest neighbors. Third, the selected M-nearest training samples should be exploited to express the original test sample and the virtual test sample and received contributions. Then it takes the advantage of the weighted sum of deviation between the two kinds of samples (test/training) to increase the facial recognition accuracy. For comparison, a face recognition method which is characteristic of simple and fast , a method which performance the nearest neighbor classifier, a novel sparse representation method named SRMVS and a two-phase method named TPTSR are applied. A plenty of experimental for face recognition using different databases reflect that our method has better classification results than the others.